技术领域Technical field
本发明属于水网渠道状态控制技术领域，尤其涉及一种基于模糊神经网络的水网渠道状态控制系统。The invention belongs to the technical field of water network channel state control, and in particular relates to a water network channel state control system based on fuzzy neural network.
背景技术Background technique
水资源与社会发展生产息息相关，水资源短缺和时空分布不均成为制约社会发展的重要因素，而水利自动化的发展能够有效提高水资源输配效率并减少灌溉渠道输水损失，从而缓解由于水资源短缺引起的社会矛盾。自20世纪30年代以来，随着自动控制技术的发展，水力自动闸门技术逐步发展，渠道自动控制技术日趋成熟，渠系自动化控制的核心是控制算法，即描述渠道输入(水位或流量误差)与输出(闸门动作)的逻辑关系的方法。Water resources are closely related to social development and production. Water shortage and uneven spatial and temporal distribution have become important factors restricting social development. The development of water conservancy automation can effectively improve the efficiency of water resources transmission and distribution and reduce water transmission losses in irrigation channels, thereby alleviating the problems caused by water resources. Social conflicts caused by shortages. Since the 1930s, with the development of automatic control technology, hydraulic automatic gate technology has gradually developed, and channel automatic control technology has become increasingly mature. The core of canal system automatic control is the control algorithm, which describes the relationship between channel input (water level or flow error) and Method of outputting (gate action) logical relationship.
随着信息技术的发展，各行业都朝着数字化、信息化、网络化、智能化的产业模式快速发展，机器学习已经成为了大多数科学中数据驱动问题的普遍和不可或缺的解决方案。但是，这些数据驱动问题在数据采集过程中固有地存在着不确定性。特别是在工程中从传感器中收集的数据，由于测量误差、知识不完整和主体差异等原因，容易面临各种不确定性。各种形式的不确定性可能会降低智能决策的有效性和准确性。With the development of information technology, various industries are rapidly developing towards digital, informatized, networked, and intelligent industrial models. Machine learning has become a common and indispensable solution to data-driven problems in most sciences. However, these data-driven problems have inherent uncertainties in the data collection process. Especially the data collected from sensors in engineering are prone to face various uncertainties due to measurement errors, incomplete knowledge, and subject differences. Various forms of uncertainty may reduce the effectiveness and accuracy of intelligent decision-making.
模糊推理系统利用模糊逻辑来度量具有不完全或不确定信息的值，已被证明是处理不确定性的强大而有效的方法。因此，模糊神经网络(FNN)将模糊推理系统和神经网络结合起来，可以很好地解决不确定大数据环境下的机器学习任务。因此，如何利用机器学习解决目前渠道控制精度不高的问题有待解决。Fuzzy inference systems utilize fuzzy logic to measure values with incomplete or uncertain information and have proven to be a powerful and effective way to deal with uncertainty. Therefore, fuzzy neural network (FNN) combines fuzzy inference system and neural network, which can well solve machine learning tasks in uncertain big data environment. Therefore, how to use machine learning to solve the current problem of low channel control accuracy remains to be solved.
发明内容Contents of the invention
针对现有技术中的上述不足，本发明提供的一种基于模糊神经网络的水网渠道状态控制系统，解决了目前渠道控制精度不高的问题。In view of the above-mentioned deficiencies in the prior art, the present invention provides a water network channel status control system based on fuzzy neural network, which solves the current problem of low channel control accuracy.
为了达到以上目的，本发明采用的技术方案为：In order to achieve the above objects, the technical solutions adopted by the present invention are:
本方案提供一种基于模糊神经网络的水网渠道状态控制系统，包括：This solution provides a water network channel status control system based on fuzzy neural network, including:
第一处理模块，用于通过水网运行的传感器获取水位数据样本；The first processing module is used to obtain water level data samples through sensors operating in the water network;
第二处理模块，用于利用fhan函数的跟踪微分器对样本进行偏差和误差修正，得到检测到的水位数据；The second processing module is used to use the tracking differentiator of the fhan function to correct the deviation and error of the sample to obtain the detected water level data;
第三处理模块，用于利用FCM聚类方法对水位数据进行处理；The third processing module is used to process water level data using the FCM clustering method;
第四处理模块，用于根据经聚类处理后的水位数据，利用RENNCOM算法训练突触权重；The fourth processing module is used to train synaptic weights using the RENNCOM algorithm based on the clustered water level data;
第五处理模块，用于根据训练得到的突触权重，设定目标函数，并设突触权重的状态方程和输出方程的结果参数，得到预测模型，所述结果参数包括每个模糊规则的参数向量；The fifth processing module is used to set the objective function according to the synaptic weight obtained by training, and set the state equation of the synaptic weight and the result parameters of the output equation to obtain a prediction model. The result parameters include the parameters of each fuzzy rule. vector;
第六处理模块，用于根据当前时刻水位，利用预测模型得到预测水位，将预测水位输入至闸门开度控制器，进行闸门开度调度，并根据水位、流量和各闸门开度控制之间的关系，以及不同工况下变量影响，形成水网渠道状态控制。The sixth processing module is used to use the prediction model to obtain the predicted water level based on the current water level, input the predicted water level to the gate opening controller, perform gate opening scheduling, and control the gate opening according to the relationship between the water level, flow rate and each gate opening. The relationship, as well as the influence of variables under different working conditions, form the state control of water network channels.
本发明的有益效果是：针对目前渠道控制精度不高问题，形成数据驱动的模糊控制，通过机理驱动控制的样本学习，经过回归模型与反馈调整，得到预测模型参数，并采用状态空间结合模糊神经网络，实现多输入多输出的状态-控制决策机制，建立状态空间，并针对输入采用模糊神经网络，实现基于状态空间模型的预测函数模糊控制。本发明解决了目前渠道控制精度不高的问题。The beneficial effects of this invention are: aiming at the current problem of low channel control accuracy, a data-driven fuzzy control is formed. Through sample learning of mechanism-driven control, the prediction model parameters are obtained through regression model and feedback adjustment, and the state space is combined with fuzzy neural network to implement a multi-input and multi-output state-control decision-making mechanism, establish a state space, and use a fuzzy neural network for the input to implement predictive function fuzzy control based on the state space model. The invention solves the current problem of low channel control accuracy.
进一步地，所述第二处理模块中，利用fhan函数的跟踪微分器对样本进行偏差和误差修正，其表达式如下：Further, in the second processing module, the tracking differentiator of the fhan function is used to correct the deviation and error of the sample, and its expression is as follows:
其中，和/>分别表示x1和x2的一阶导数，x1和x2分别表示跟踪信号和微分信号，fhan(·)表示fhan函数，v表示输入信号，r0表示速度因子，h0表示滤波因子；in, and/> represent the first-order derivatives of x1 and x2 respectively, x1 and x2 represent the tracking signal and the differential signal respectively, fhan(·) represents the fhan function, v represents the input signal, r0 represents the speed factor, and h0 represents the filter factor;
离散形式如下：The discrete form is as follows:
其中，x1(k')表示k'时刻的跟踪信号，v(k')表示k'时刻输入信号，x2(k')表示k'时刻的微分信号，x1(k'+1)表示k'+1时刻的跟踪信号，h表示滤波因子，x2(k'+1)表示k'+1时刻的微分信号；Among them, x1 (k') represents the tracking signal at time k', v (k') represents the input signal at time k', x2 (k') represents the differential signal at time k', x1 (k'+1) represents the tracking signal at k'+1 time, h represents the filter factor, x2 (k'+1) represents the differential signal at k'+1 time;
其fhan(x1(k')-v(k'),x2(k'),r0,h0)的具体形式为：The specific form of fhan(x1 (k')-v(k'),x2 (k'),r0 ,h0 ) is:
其中，d、a0、a1、a2、a、y、sy和sa均表示跟踪微分器可调参数。Among them, d, a0 , a1 , a2 , a, y, sy and sa all represent the tracking differentiator adjustable parameters.
上述进一步方案的有益效果是：本发明通过对样本进行偏差和误差修正处理，减少了数据采集过程中对水位数据的干扰，得到准确的水位数据果。The beneficial effects of the above further solution are: by performing deviation and error correction processing on the samples, the present invention reduces the interference to the water level data during the data collection process and obtains accurate water level data results.
再进一步地，所述利用FCM聚类方法对水位数据进行处理，其表达式如下：Furthermore, the FCM clustering method is used to process the water level data, and its expression is as follows:
其中，σij表示隶属函数的标准差，P表示水位数据样本集总数，uin表示隶属，r表示聚类总数，mi'j表示m维水位样本数据集中的第i'行第j列数据，xj'(n)表示第j'个输入向量，mkj表示m维水位样本数据集中的第k'行第j列数据，uin(n)表示第n个水位数据样本属于第i个聚类的隶属度，c表示模糊性参数数。Among them, σij represents the standard deviation of the membership function, P represents the total number of water level data sample sets, uin represents membership, r represents the total number of clusters, mi'j represents the i'th row and j column data in the m-dimensional water level sample data set. , xj' (n) represents the j'th input vector, mkj represents the k'th row and j column data in the m-dimensional water level sample data set, uin (n) represents the nth water level data sample belonging to the i-th The membership degree of clustering, c represents the number of fuzziness parameters.
上述进一步方案的有益效果是：本发明利用FCM聚类方法对水位数据进行处理，收敛速度快，能够快速对大规模数据进行聚类，对设备算力要求低。The beneficial effects of the above further solution are: the present invention uses the FCM clustering method to process water level data, has a fast convergence speed, can quickly cluster large-scale data, and has low requirements on equipment computing power.
再进一步地，所述第四处理模块包括：Furthermore, the fourth processing module includes:
第一处理单元，用于通过以下表达式确定N个神经元的突触权重BDRNN输出：The first processing unit is used to determine the synaptic weight BDRNN output of N neurons through the following expression:
其中，和/>分别表示第i”条模糊规则下处理第n个水位数据样本时形成的第k-1和第k个的神经元输出，f1表示激活函数，j”表示输入向理的维数指标，m表示水位样本数据集的维度，/>和/>分别表示第i”条模糊规则下构成第k-1和第k个的神经元的突触权值，xj'(n)表示第j'个输入向量，/>和/>均表示第i”条模糊规则下突触权重BDRNN第k个的神经元的反馈权值，/>表示第i”条模糊规则下处理第n-1个水位数据样本时形成的第k-1个的神经元，/>表示第i”条模糊规则下处理第n-1个水位数据样本时形成的第k个的神经元，r表示聚类总数，其第i”条模糊规则对应第i个聚类，N表示第N个神经元；in, and/> Respectively represent the k-1 and k-th neuron outputs formed when processing the n-th water level data sample under the i"-th fuzzy rule, f1 represents the activation function, j" represents the dimension index of the input theory, m Represents the dimension of the water level sample data set,/> and/> represent the synaptic weights of the k-1 and k-th neurons respectively under the i"-th fuzzy rule, xj' (n) represents the j'-th input vector,/> and/> Both represent the feedback weight of the k-th neuron of the synaptic weight BDRNN under the i"-th fuzzy rule,/> Represents the k-1th neuron formed when processing the n-1th water level data sample under the i"-th fuzzy rule,/> represents the k-th neuron formed when processing the n-1 water level data sample under the i"-th fuzzy rule, r represents the total number of clusters, and the i"-th fuzzy rule corresponds to the i-th cluster, and N represents the N neurons;
第二处理单元，用于根据确定的突触权重BDRNN输出，对于第i”条模糊规则，计算得到突触权重BDRNN的输出：The second processing unit is used to output the synaptic weight BDRNN based on the determined synaptic weight. For the i" fuzzy rule, calculate the output of the synaptic weight BDRNN:
其中，gi”(n)表示第i”条模糊规则下的突触权重BDRNN输出，f2表示激活函数，bij表示输出神经元的突触权值，表示第i”条模糊规则下处理第n个水位数据样本时形成的第k块的神经元输出；Among them, gi" (n) represents the synaptic weight BDRNN output under the i"-th fuzzy rule, f2 represents the activation function, bij represents the synaptic weight of the output neuron, Represents the neuron output of the k-th block formed when processing the n-th water level data sample under the i-th fuzzy rule;
第三处理单元，用于根据第五处理单元得到的突触权重BDRNN的输出，将每个区块的反馈权重通过以下反馈矩阵组成，完成对突触权重的训练：The third processing unit is used to combine the feedback weight of each block with the following feedback matrix based on the output of the synaptic weight BDRNN obtained by the fifth processing unit to complete the training of the synaptic weight:
其中，表示第i”条模糊规则下的反馈矩阵。in, Represents the feedback matrix under the i" fuzzy rule.
上述进一步方案的有益效果是：本发明通过训练突触权重，可以纳入关于学习过程的各种约束。在这种情况下，基于梯度下降的方法不能保证稳定学习，因为在BDRNN的隐层存在反馈连接。因此，可以以适当的函数形式引入与稳定学习相关的约束，并与标准误差函数同时进行优化。The beneficial effect of the above further solution is that the present invention can incorporate various constraints on the learning process by training synaptic weights. In this case, gradient descent-based methods cannot guarantee stable learning because there are feedback connections in the hidden layers of BDRNN. Therefore, constraints related to stable learning can be introduced in an appropriate functional form and optimized simultaneously with the standard error function.
再进一步地，所述反馈矩阵的特征值的约束条件如下：Furthermore, the constraints on the eigenvalues of the feedback matrix are as follows:
其中，λi”k表示第i”条模糊规则下的的特征值，/>和/>均表示第i”条模糊规则下突触权重BDRNN第k个的神经元的反馈权值。Among them, λi”k represents the i”th fuzzy rule eigenvalues,/> and/> Both represent the feedback weight of the k-th neuron of the synaptic weight BDRNN under the i"-th fuzzy rule.
再进一步地，所述预测模型的目标函数的表达式如下：Furthermore, the expression of the objective function of the prediction model is as follows:
其中，pik和fd(zik)均表示预测模型的目标函数，e表示自然对数，as表示sigmoid函数的斜率，λi”k表示第i”条模糊规则下的的特征值，/>和/>均表示第i”条模糊规则下突触权重BDRNN第k个的神经元的反馈权值。Among them, pik and fd (zik ) both represent the objective function of the prediction model, e represents the natural logarithm, as represents the slope of the sigmoid function, and λi”k represents the i”th fuzzy rule. eigenvalues,/> and/> Both represent the feedback weight of the k-th neuron of the synaptic weight BDRNN under the i"-th fuzzy rule.
再进一步地，所述目标函数的约束条件为：Furthermore, the constraints of the objective function are:
(1)设定误差最小化，采用均方差MSE为度量：(1) Set the error to be minimized and use the mean square error MSE as the metric:
其中，P表示水位数据样本集总数，n'表示时间，y(n)和分别表示水位和流量的时间序列；Among them, P represents the total number of water level data sample sets, n' represents time, y(n) and represent the time series of water level and flow respectively;
(2)设定补偿函数Φ最小化：(2) Set the compensation function Φ to minimize:
其中，r表示表示聚类总数，i表示聚类个数，N表示水位时间序列数据总数，k表示神经元个数，pik表示稳定性函数，和/>分别表示第i”条模糊规则下突触权重BDRNN第k个的神经元的反馈权值；Among them, r represents the total number of clusters, i represents the number of clusters, N represents the total number of water level time series data, k represents the number of neurons, pik represents the stability function, and/> Respectively represent the feedback weight of the k-th neuron of the synaptic weight BDRNN under the i"-th fuzzy rule;
(3)设定权重空间搜索条件Φw：(3) Set the weight space search condition Φw :
Φw＝dθT·(Δ2)-1·dθ-1＝0Φw ＝dθT ·(Δ2 )-1 ·dθ-1＝0
其中，dθ表示每次迭代时结果参数向量的变化量，Δ表示包含每个权重最大参数变化的对角矩阵，T表示转置。Among them, dθ represents the change amount of the result parameter vector in each iteration, Δ represents the diagonal matrix containing the maximum parameter change of each weight, and T represents the transpose.
上述进一步方案的有益效果是：本发明通过设定的三个目标，能够使得最后得到的预测模型具有误差最小的特性。The beneficial effect of the above further solution is that by setting three goals, the present invention can make the final prediction model have the characteristics of minimum error.
附图说明Description of the drawings
图1为本发明的系统结构示意图。Figure 1 is a schematic diagram of the system structure of the present invention.
具体实施方式Detailed ways
下面对本发明的具体实施方式进行描述，以便于本技术领域的技术人员理解本发明，但应该清楚，本发明不限于具体实施方式的范围，对本技术领域的普通技术人员来讲，只要各种变化在所附的权利要求限定和确定的本发明的精神和范围内，这些变化是显而易见的，一切利用本发明构思的发明创造均在保护之列。The specific embodiments of the present invention are described below to facilitate those skilled in the art to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the technical field, as long as various changes These changes are obvious within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions and creations utilizing the concept of the invention are protected.
实施例Example
如图1所示，本发明提供了一种基于模糊神经网络的水网渠道状态控制系统，包括：As shown in Figure 1, the present invention provides a water network channel status control system based on fuzzy neural network, including:
第一处理模块，用于通过水网运行的传感器获取水位数据样本First processing module for obtaining water level data samples from sensors operating on the water network
本实施例中，选定水网运行的传感器数据作为样本，并将每一个样本中在同一时刻多个点位的水位值作为自变量，将与每一个样本对应的闸门控制指引作为因变量；自变量为该系统的输入值，即输入至第一模块的值；因变量为该系统的最终输出值，即第六模块处理完成后的输出的闸门开度控制。In this embodiment, the sensor data of the water network operation is selected as the sample, the water level values of multiple points in each sample at the same time are used as the independent variables, and the gate control guidance corresponding to each sample is used as the dependent variable; The independent variable is the input value of the system, that is, the value input to the first module; the dependent variable is the final output value of the system, that is, the gate opening control of the output after the processing of the sixth module is completed.
本实施例中，选定水网运行的传感器数据作为样本，进行监督学习，一方面从回归模型解决预测问题，将每一个样本中控制过程在同一时刻多点的水位值作为自变量，将与其对应的闸门控制指引作为因变量，可运用的回归分析方法包含但不限于：LinearRegression线性回归、Logistic Regression逻辑回归、Polynomial Regression多项式回归、Stepwise Regression逐步回归、Ridge Regression岭回归、Lasso Regression套索回归以及Elastic Net回归等。In this embodiment, the sensor data of the water network operation is selected as a sample to perform supervised learning. On the one hand, the prediction problem is solved from the regression model, and the water level values of multiple points of the control process at the same time in each sample are used as independent variables. The corresponding gate control guideline is used as the dependent variable. The applicable regression analysis methods include but are not limited to: LinearRegression, Logistic Regression, Polynomial Regression, Stepwise Regression, Ridge Regression, Lasso Regression, and Elastic Net regression, etc.
第二处理模块，用于利用fhan函数的跟踪微分器对样本进行偏差和误差修正，得到检测到的水位数据；The second processing module is used to use the tracking differentiator of the fhan function to correct the deviation and error of the sample to obtain the detected water level data;
本实施例中，本发明引入基于fhan函数的跟踪微分器，运用其跟踪和滤波功能，解决误差和噪音等问题，进一步运用输出信号无超调情况下快速跟踪原始信号，形成观测器的闭环控制，形成有利于学习的样本。计算采用算法如下：In this embodiment, the present invention introduces a tracking differentiator based on the fhan function, uses its tracking and filtering functions to solve problems such as errors and noise, and further uses the output signal to quickly track the original signal without overshoot, forming a closed-loop control of the observer. , forming samples that are conducive to learning. The calculation algorithm is as follows:
离散形式如下：The discrete form is as follows:
其中，和/>分别表示x1和x2的一阶导数，x1和x2分别表示跟踪信号和微分信号，fhan(·)表示fhan函数，v表示输入信号，r0表示速度因子，h0表示滤波因子，fh表示fhan函数，x1(k')表示k'时刻的跟踪信号，v(k')表示k'时刻输入信号，x2(k')表示k'时刻的微分信号，x1(k'+1)表示k'+1时刻的跟踪信号，h表示滤波因子，x2(k'+1)表示k'+1时刻的微分信号。in, and/> represent the first-order derivatives of x1 and x2 respectively, x1 and x2 represent the tracking signal and the differential signal respectively, fhan(·) represents the fhan function, v represents the input signal, r0 represents the speed factor, h0 represents the filter factor, fh represents the fhan function, x1 (k') represents the tracking signal at time k', v (k') represents the input signal at time k', x2 (k') represents the differential signal at time k', x1 (k' +1) represents the tracking signal at k'+1 time, h represents the filter factor, and x2 (k'+1) represents the differential signal at k'+1 time.
其fhan(x1(k')-v(k'),x2(k'),r0,h0)具体形式：Its specific form: fhan(x1 (k')-v(k'),x2 (k'),r0 ,h0 ):
其中，d、a0、a1、a2、a、y、sy和sa均表示跟踪微分器可调参数；Among them, d, a0 , a1 , a2 , a, y, sy and sa all represent the adjustable parameters of the tracking differentiator;
选择合理的参数，准确运用跟踪微分器，对机理驱动的联动控制过程中产生的偏差和误差做出修正得到最终检测到的水位。Select reasonable parameters, accurately use the tracking differentiator, and correct the deviations and errors generated during the mechanism-driven linkage control process to obtain the final detected water level.
第三处理模块，用于利用FCM聚类方法对水位数据进行处理；The third processing module is used to process water level data using the FCM clustering method;
本实施例中，采用FCM聚类方法对数据样本水位数据进行处理，根据属于每个聚类的数据样本之间最小距离提供最合适的聚类中心，是的每一个聚类中心都集中在数据集的一部分上，对于给定数量的聚类和样本的数据集，聚类中心表示如下式：In this embodiment, the FCM clustering method is used to process the data sample water level data, and the most appropriate clustering center is provided based on the minimum distance between the data samples belonging to each cluster, so that each clustering center is concentrated in the data On a part of the set, for a data set with a given number of clusters and samples, the cluster center is expressed as follows:
进行隶属函数的标准差计算以确定模糊规则的前提部分保持不变，计算公式如下：The standard deviation of the membership function is calculated to determine that the premise of the fuzzy rule remains unchanged. The calculation formula is as follows:
其中，σij表示隶属函数的标准差，P表示水位数据样本集总数，uin表示隶属，r表示聚类总数，mi'j表示m维水位样本数据集中的第i'行第j列数据，xj'(n)表示第j'个输入向量，mkj表示m维水位样本数据集中的第k'行第j列数据，uin(n)表示第n个水位数据样本属于第i个聚类的隶属度，c表示模糊性参。Among them, σij represents the standard deviation of the membership function, P represents the total number of water level data sample sets, uin represents membership, r represents the total number of clusters, mi'j represents the i'th row and j column data in the m-dimensional water level sample data set. , xj' (n) represents the j'th input vector, mkj represents the k'th row and j column data in the m-dimensional water level sample data set, uin (n) represents the nth water level data sample belonging to the i-th The membership degree of clustering, c represents the fuzziness parameter.
第四处理模块，用于根据经聚类处理后的水位数据，利用RENNCOM算法训练突触权重，包括：The fourth processing module is used to train synaptic weights using the RENNCOM algorithm based on the clustered water level data, including:
第一处理单元，用于通过以下表达式确定N个神经元的突触权重BDRNN输出：The first processing unit is used to determine the synaptic weight BDRNN output of N neurons through the following expression:
其中，和/>分别表示第i”条模糊规则下处理第n个水位数据样本时形成的第k-1和第k个的神经元输出，f1表示激活函数，j”表示输入向理的维数指标，m表示水位样本数据集的维度，/>和/>分别表示第i”条模糊规则下构成第k-1和第k个的神经元的突触权值，xj'(n)表示第j'个输入向量，/>和/>均表示第i”条模糊规则下突触权重BDRNN第k个的神经元的反馈权值，/>表示第i”条模糊规则下处理第n-1个水位数据样本时形成的第k-1个的神经元，/>表示第i”条模糊规则下处理第n-1个水位数据样本时形成的第k个的神经元，r表示聚类总数，其第i”条模糊规则对应第i个聚类，N表示第N个神经元；in, and/> Respectively represent the k-1 and k-th neuron outputs formed when processing the n-th water level data sample under the i"-th fuzzy rule, f1 represents the activation function, j" represents the dimension index of the input theory, m Represents the dimension of the water level sample data set,/> and/> represent the synaptic weights of the k-1 and k-th neurons respectively under the i"-th fuzzy rule, xj' (n) represents the j'-th input vector,/> and/> Both represent the feedback weight of the k-th neuron of the synaptic weight BDRNN under the i"-th fuzzy rule,/> Represents the k-1th neuron formed when processing the n-1th water level data sample under the i"-th fuzzy rule,/> represents the k-th neuron formed when processing the n-1 water level data sample under the i"-th fuzzy rule, r represents the total number of clusters, and the i"-th fuzzy rule corresponds to the i-th cluster, and N represents the N neurons;
第二处理单元，用于根据确定的突触权重BDRNN输出，对于第i”条模糊规则，计算得到突触权重BDRNN的输出：The second processing unit is used to output the synaptic weight BDRNN based on the determined synaptic weight. For the i" fuzzy rule, calculate the output of the synaptic weight BDRNN:
其中，gi”(n)表示第i”条模糊规则下的突触权重BDRNN输出，f2表示激活函数，bij表示输出神经元的突触权值，表示第i”条模糊规则下处理第n个水位数据样本时形成的第k块的神经元输出；Among them, gi" (n) represents the synaptic weight BDRNN output under the i"-th fuzzy rule, f2 represents the activation function, bij represents the synaptic weight of the output neuron, Represents the neuron output of the k-th block formed when processing the n-th water level data sample under the i-th fuzzy rule;
第三处理单元，用于根据第五处理单元得到的突触权重BDRNN的输出，将每个区块的反馈权重通过以下反馈矩阵组成，完成对突触权重的训练：The third processing unit is used to combine the feedback weight of each block with the following feedback matrix based on the output of the synaptic weight BDRNN obtained by the fifth processing unit to complete the training of the synaptic weight:
其中，表示第i”条模糊规则下的反馈矩阵。in, Represents the feedback matrix under the i" fuzzy rule.
第五处理模块，用于根据训练得到的突触权重，设定目标函数，并设突触权重的状态方程和输出方程的结果参数，所述结果参数包括每个模糊规则的参数向量；The fifth processing module is used to set the objective function according to the synaptic weight obtained by training, and set the state equation of the synaptic weight and the result parameters of the output equation, where the result parameters include the parameter vector of each fuzzy rule;
本实施例中，设定目标函数。适合并入约束方程的目标函数是sigmoid函数，它是光滑且连续可微的，因此对寄生振荡引起的问题具有鲁棒性。因此，预测模型的目标函数的表达式如下：In this embodiment, an objective function is set. An objective function suitable for incorporation into the constraint equation is the sigmoid function, which is smooth and continuously differentiable and therefore robust to problems caused by spurious oscillations. Therefore, the expression of the objective function of the prediction model is as follows:
其中，pik和fd(zik)均表示预测模型的目标函数，e表示自然对数，as表示sigmoid函数的斜率，λi”k表示第i”条模糊规则下的的特征值，/>和/>均表示第i”条模糊规则下突触权重BDRNN第k个的神经元的反馈权值。当特征值在单位圆内时，模型是稳定的，因此其活动区域as的取值在[4,8]范围内。Among them, pik and fd (zik ) both represent the objective function of the prediction model, e represents the natural logarithm, as represents the slope of the sigmoid function, and λi”k represents the i”th fuzzy rule. eigenvalues,/> and/> Both represent the feedback weight of the k-th neuron of the synaptic weight BDRNN under the i" fuzzy rule. When the eigenvalue is within the unit circle, the model is stable, so the value of its active area as is in [4, 8] within the range.
本实施例中，设BDRNN的状态方程和输出方程结果参数包括每个模糊规则的参数向量θcon。利用RENNCOM算法实现目标函数的以下三个目标：In this embodiment, it is assumed that the state equation and output equation result parameters of BDRNN include the parameter vector θcon of each fuzzy rule. The RENNCOM algorithm is used to achieve the following three goals of the objective function:
(1)设定误差最小化，采用均方差MSE为度量：(1) Set the error to be minimized and use the mean square error MSE as the metric:
其中，P表示水位数据样本集总数，n'表示时间，y(n)和分别表示水位和流量的时间序列；Among them, P represents the total number of water level data sample sets, n' represents time, y(n) and represent the time series of water level and flow respectively;
(2)设定补偿函数Φ最小化：(2) Set the compensation function Φ to minimize:
其中，r表示表示聚类总数，i表示聚类个数，N表示水位时间序列数据总数，k表示神经元个数，pik表示稳定性函数，和/>分别表示第i”条模糊规则下突触权重BDRNN第k个的神经元的反馈权值；Among them, r represents the total number of clusters, i represents the number of clusters, N represents the total number of water level time series data, k represents the number of neurons, pik represents the stability function, and/> Respectively represent the feedback weight of the k-th neuron of the synaptic weight BDRNN under the i"-th fuzzy rule;
(3)设定权重空间搜索条件Φw：(3) Set the weight space search condition Φw :
Φw＝dθT·(Δ2)-1·dθ-1＝0Φw ＝dθT ·(Δ2 )-1 ·dθ-1＝0
其中，dθ表示每次迭代时结果参数向量的变化量，Δ表示包含每个权重最大参数变化的对角矩阵，T表示转置。Among them, dθ represents the change amount of the result parameter vector in each iteration, Δ represents the diagonal matrix containing the maximum parameter change of each weight, and T represents the transpose.
第六处理模块，用于根据第五处理模块的处理结果，将输出的预测水位输入至闸门开度控制器，进行闸门开度调度，并根据水位、流量和各闸门开度控制之间的关系，以及不同情形下变量影响，形成水网渠道状态控制。The sixth processing module is used to input the output predicted water level to the gate opening controller according to the processing results of the fifth processing module, perform gate opening scheduling, and control the gate opening according to the relationship between the water level, flow rate and each gate opening. , and the influence of variables under different situations, forming water network channel status control.
本实施例中，将输出的预测水位输入至闸门开度控制器进行闸门开度调度，并根据水位、流量和各个闸门开度控制之间的关系，以及不同工况下的变量影响，并进行检验、反馈、调整、完善，最终实现运用。In this embodiment, the output predicted water level is input to the gate opening controller for gate opening scheduling, and based on the relationship between water level, flow rate and each gate opening control, as well as the influence of variables under different working conditions, and Inspection, feedback, adjustment, improvement, and finally implementation.
本实施例中，本发明通过以上设计，最终形成基于规则的模糊控制，制定输入与输出之间的规则，根据输入，通过规则判断得到输出值，实现输入水位值，输出控制指引，最后实现基于预测函数控制算法原理设计基于状态空间模型的预测函数控制器，解决状态一旦改变，如流量的大小、水位的高低、需水目标的设置，以前的学习不能适用的问题。实现不同工况下得出的准确的控制操作输出。In this embodiment, through the above design, the present invention finally forms a rule-based fuzzy control, formulates the rules between input and output, according to the input, obtains the output value through rule judgment, realizes the input water level value, outputs the control guide, and finally realizes the based on The principle of predictive function control algorithm is to design a predictive function controller based on the state space model to solve the problem that once the state changes, such as the size of the flow, the level of the water level, and the setting of the water demand target, the previous learning cannot be applied. Achieve accurate control operation output under different working conditions.
本实施例中，本发明提出了在模糊控制在供水-发电渠道中的应用创新。形成供水-发电渠道联动控制的多输入多输出状态空间，将同一时刻所有测点的水位作为多输入得到的全部闸门的开度控制运用指引作为多输出，确保不同工况之间控制的高效平稳衔接。创新应用深度学习达成优化方法和运行方略，并引入模糊神经网络，能够对控制对象进行分层分类，进一步从状态空间得到正确合理的决策指引，形成基于预测函数控制算法原理设计基于状态空间模型的预测函数控制器。In this embodiment, the present invention proposes an innovation in the application of fuzzy control in the water supply-power generation channel. Form a multi-input multi-output state space for water supply-power generation channel linkage control. The water levels at all measuring points at the same time are used as multi-inputs and the opening control application guidelines for all gates are used as multi-outputs to ensure efficient and smooth control between different working conditions. connection. Innovative application of deep learning to achieve optimization methods and operation strategies, and the introduction of fuzzy neural networks, which can hierarchically classify control objects, further obtain correct and reasonable decision-making guidance from the state space, and form a state-space model based on the principle of predictive function control algorithm. Predictive function controller.
本实施例中，本发明提出供水-发电渠道控制监测中的抗扰动量测创新。对渠道量测装置的初始数据，运用抗干扰和滤波算法，避免误差，反馈真实准确。进一步提高了准确性，能够将准确的实时水位和流量反馈至调节器中，支撑联动控制。In this embodiment, the present invention proposes anti-disturbance measurement innovation in water supply-power generation channel control and monitoring. For the initial data of the channel measurement device, anti-interference and filtering algorithms are used to avoid errors and the feedback is true and accurate. The accuracy is further improved, and accurate real-time water level and flow can be fed back to the regulator to support linkage control.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310945481.2ACN116880201A (en) | 2023-07-28 | 2023-07-28 | Water network channel status control system based on fuzzy neural network |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310945481.2ACN116880201A (en) | 2023-07-28 | 2023-07-28 | Water network channel status control system based on fuzzy neural network |
| Publication Number | Publication Date |
|---|---|
| CN116880201Atrue CN116880201A (en) | 2023-10-13 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310945481.2APendingCN116880201A (en) | 2023-07-28 | 2023-07-28 | Water network channel status control system based on fuzzy neural network |
| Country | Link |
|---|---|
| CN (1) | CN116880201A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117824788A (en)* | 2024-03-05 | 2024-04-05 | 河海大学 | Water level monitoring and analysis system |
| CN119168427A (en)* | 2024-11-20 | 2024-12-20 | 杭州定川信息技术有限公司 | Irrigation canal network water distribution optimization control method and device, electronic equipment, medium |
| CN119335879A (en)* | 2024-12-19 | 2025-01-21 | 北京慧图科技(集团)股份有限公司 | Automatic Regulation System of Water Gate in Irrigation Area Based on Fuzzy Logic Control |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117824788A (en)* | 2024-03-05 | 2024-04-05 | 河海大学 | Water level monitoring and analysis system |
| CN117824788B (en)* | 2024-03-05 | 2024-05-28 | 河海大学 | Water level monitoring and analyzing system |
| CN119168427A (en)* | 2024-11-20 | 2024-12-20 | 杭州定川信息技术有限公司 | Irrigation canal network water distribution optimization control method and device, electronic equipment, medium |
| CN119335879A (en)* | 2024-12-19 | 2025-01-21 | 北京慧图科技(集团)股份有限公司 | Automatic Regulation System of Water Gate in Irrigation Area Based on Fuzzy Logic Control |
| Publication | Publication Date | Title |
|---|---|---|
| Guo et al. | A self-interpretable soft sensor based on deep learning and multiple attention mechanism: From data selection to sensor modeling | |
| JP7346705B2 (en) | Water level prediction and control method for series water channel based on fuzzy neural network | |
| CN116880201A (en) | Water network channel status control system based on fuzzy neural network | |
| CN109407654B (en) | A nonlinear causal analysis method for industrial data based on sparse deep neural network | |
| CN111353631A (en) | Prediction method of vacuum degree of thermal power plant condenser based on multi-layer LSTM | |
| Tian et al. | Time-delay compensation method for networked control system based on time-delay prediction and implicit PIGPC | |
| CN110245398B (en) | Soft measurement deep learning method for thermal deformation of air preheater rotor | |
| CN112434848A (en) | Nonlinear weighted combination wind power prediction method based on deep belief network | |
| CN113325721A (en) | Model-free adaptive control method and system for industrial system | |
| CN108537366A (en) | Reservoir scheduling method based on optimal convolution bidimensionalization | |
| Zhou et al. | Deep learning with PID residual elimination network for time-series prediction of water quality in aquaculture industry | |
| CN104050505A (en) | Multilayer-perceptron training method based on bee colony algorithm with learning factor | |
| CN113962454A (en) | LSTM energy consumption prediction method based on dual feature selection + particle swarm optimization | |
| Sun et al. | Evaluation of university project based on partial least squares and dynamic back propagation neural network group | |
| CN120234763B (en) | A wind speed and direction prediction method based on improved TCN-LSTM | |
| CN117440017A (en) | Flow big data acquisition and intelligent monitoring method and Internet of things system thereof | |
| Zeng et al. | DDPG-based continuous thickness and tension coupling control for the unsteady cold rolling process | |
| Wang et al. | Application of artificial neural networks in chemical process control | |
| Wang et al. | How deep is deep enough for deep belief network for approximating model predictive control law | |
| Tian et al. | Generalized predictive PID control for main steam temperature based on improved PSO algorithm | |
| CN114415503B (en) | Temperature big data IoT detection and intelligent control system | |
| Yang et al. | Application of fuzzy neural network PID algorithm in oil pump control | |
| CN116865343B (en) | Model-free adaptive control method, device and medium for distributed photovoltaic distribution network | |
| CN118732605A (en) | An intelligent control system for tool stabilization device of multi-axis linkage of CNC machine tools based on LSTM neural network and PID optimization | |
| Juan et al. | Multi-step prediction of dissolved oxygen in river based on random forest missing value imputation and attention mechanism coupled with recurrent neural network |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |