Induction Algorithm Research Articles

Overview of Fundamental Frequency Sensorless Algorithms for AC Motors: A Unified Perspective

This article uses the active flux concept to review fundamental frequency sensorless algorithms for both induction and permanent magnet motors in one framework. Fundamentally, sensorless torque estimation can be directly solved using voltage-current model (VM) estimator or indirectly solved using current-speed model (CM) estimator. The latter turns the torque estimation problem into a speed estimation problem. The stator flux in VM and the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$d$ </tex-math></inline-formula> -axis angle in CM are deemed as the two sets of original states for the sensorless drive. Through the change of states, the direct torque estimation can be realized via observer designs, whereas the speed dependency in the dynamics of the unknown state [e.g., active flux and electromotive force (emf)] gives rise to a class of speed estimation methods, known as model reference adaptive system (MRAS). The idea of a general speed observer is proposed to summarize various separate speed estimation methods needed for direct torque estimation. It is suggested to adopt inherently sensorless designs such that two-way coupling between torque estimation and speed estimation is avoided. For induction motors, it turns out that the unmodeled voltage in the active flux dynamics reveals current flowing in rotor bars and can be further modeled, for which the solutions to regeneration instability problem are discussed, and change of states is recommended to attain global stability. Finally discussed are the results of the slow reversal test, where local weak observability of ac motors can be potentially preserved.

Analysis of the Impact of Vibrations on a Micro-Hydraulic Valve Using a Modified Induction Algorithm

This paper addresses the impact of mechanical vibrations of different frequencies on a particular type of valve. It has been shown that a neural network can be used to compress measurement data and determine the frequency range that is most important in describing the impact of mechanical vibrations on a micro-hydraulic overflow valve. Later, induction decision trees were used for the generated areas, determining key measurement points. The most important areas of dependence are determined using inductive decision trees in induction. The entropy measure is used to determine the most significant attribute. A modified induction algorithm was used for the comprehensive analyses. The analysis carried out in the paper identified the intervals in which the flow rate plays a decisive role for the entire amplitude and frequency spectrum. An analysis was performed for the 200…900 [Hz] frequency interval of the external driving force, with a harmonic step of 10 [Hz]. The analysis was performed while considering these main valve parameters: the pressure of the overflow valve opening p = 10 [MPa]; flow rate in the valve: 0.6, 0.8 and 1 [dm3/min]; stiffness of valve spring c = 7.49 [N/mm]. Plots were presented and for each plot, the most important four intervals were determined. They are hierarchically ordered in the interval range of the whole frequency spectrum range: (coefficient IV)—the most important; (coefficient III)—important; (coefficient II)—less important; (coefficient I)—least important. A test rig and the results of a study on the effect of mechanical vibration on changes in the amplitude-frequency spectrum of pressure pulsations of a micro-hydraulic system, in which a micro-hydraulic relief valve was subjected to mechanical vibration, are presented.

RIDDLE: Rule Induction with Deep Learning

Numerous applications rely on the efficiency of Deep Learning models to address complex classification tasks for critical decisions-making. However, we may not know how each feature in these models contributes towards the prediction. In contrast, Rule Induction algorithms provide an interpretable way to extract patterns from data, but traditional approaches suffer in terms of scalability. In this work, we bridge Deep Learning and Rule Induction and define the RIDDLE (Rule Induction with Deep Learning) architecture. We show that RIDDLE has state-of-the-art performance in Rule Induction via an empirical evaluation.

A pivot-based simulated annealing algorithm to determine oblique splits for decision tree induction

We describe a new simulated annealing algorithm to compute near-optimal oblique splits in the context of decision tree induction. The algorithm can be interpreted as a walk on the cells of a hyperplane arrangement defined by the observations in the training set. The cells of this hyperplane arrangement correspond to subsets of oblique splits that divide the feature space in the same manner and the vertices of this arrangement reveal multiple neighboring solutions. We use a pivoting strategy to iterate over the vertices and to explore this neighborhood. Embedding this neighborhood search in a simulated annealing framework allows to escape local minima and increases the probability of finding global optimal solutions. To overcome the problems related to degeneracy, we rely on a lexicographic pivoting scheme. Our experimental results indicate that our approach is well-suited for inducing small and accurate decision trees and capable of outperforming existing univariate and oblique decision tree induction algorithms. Furthermore, oblique decision trees obtained with this method are competitive with other popular prediction models.

Correction to: Modeling landslide susceptibility using data mining techniques of kernel logistic regression, fuzzy unordered rule induction algorithm, SysFor and random forest

Correction to: Modeling landslide susceptibility using data mining techniques of kernel logistic regression, fuzzy unordered rule induction algorithm, SysFor and random forest

Two Approaches for Option Pricing under Illiquidity

The paper focuses on option pricing under unusual behaviour of the market, when the price may not be changed for some time what is quite a common situation on the modern financial markets. There are some patterns that can cause permanent price gaps to form and lead to illiquidity. For example, global changes that have a negative impact on financial activity, or a small number of market participants, or the market is quite young and is just in the process of developing, etc.In the paper discrete and continuous time approaches for modelling market with illiquidity and evaluation option pricing were considered.Trinomial discrete time model improves upon the binomial model by allowing a stock price not only to move up, down but stay the same with certain probabilities, what is a desirable feature for the illiquid modelling. In the paper parameters for real financial data were identified and the backward induction algorithm for building call option price trinomial tree was applied.Subdiffusive continuous time model allows successfully apply the physical models for describing the trapping events to model financial data stagnation's periods. In this paper the Inverse Gaussian process IG was proposed as a subordinator for the subdiffusive modelling of illiquidity and option pricing. The simulation of the trajectories for subordinator, inverse subordinator and subdiffusive GBM were performed. The Monte Carlo method for option evaluation was applied.Our aim was not only to compare these two models each with other, but also to show that both models adequately describe the illiquid market and can be used for option pricing on this market. For this purpose absolute relative percentage (ARPE) and root mean squared error (RMSE) for both models were computed and analysed.Thanks to the proposed approaches, the investor gets a tools, which allows him to take into account the illiquidity.

Optimal utilization of integrated photovoltaic battery systems: An application in the residential sector

PhotoVoltaic (PV) panels have been increasingly favored by residential users in recent years, due to noticeable reductions in their costs. The PV systems become more effective when combined with battery packages, which store the energy produced by the PV systems for later use. This way, the PV systems are able to provide flexible and reliable services even when the peak demand for electricity misalign with the window of most efficient PV power generation. In this study, we develop an integrated charge/discharge scheme for lithium-ion batteries to maximize their total expected benefit. Specifically, we develop a Markov Decision Process (MDP) model to maximize the battery utilization, subject to uncertainty in weather conditions and electricity demands, while accounting for battery degradation due to calendar aging and charging/discharging cycles. Due to the extremely slow rate of degradation in batteries, the state space of the MDP is excessively large. To solve the problem efficiently, we establish structural properties of the MDP and exploit them to solve the problem. We improve the backward induction algorithm with established structural properties. We further improve the Deep Q-network (DQN) algorithm by proposing two novel algorithms, the augmented DQN (ADQN) algorithm and the stochastic augmented DQN (SADQN) algorithm. Computational results indicate that ADQN and SADQN solve the problem much faster than DQN, with better solution qualities. The ADQN and SADQN algorithms provide flexibility for practitioners in real-world implementations.

Identifying circRNA-miRNA interaction based on multi-biological interaction fusion.

CircRNA is a new type of non-coding RNA with a closed loop structure. More and more biological experiments show that circRNA plays important roles in many diseases by regulating the target genes of miRNA. Therefore, correct identification of the potential interaction between circRNA and miRNA not only helps to understand the mechanism of the disease, but also contributes to the diagnosis, treatment, and prognosis of the disease. In this study, we propose a model (IIMCCMA) by using network embedding and matrix completion to predict the potential interaction of circRNA-miRNA. Firstly, the corresponding adjacency matrix is constructed based on the experimentally verified circRNA-miRNA interaction, circRNA-cancer interaction, and miRNA-cancer interaction. Then, the Gaussian kernel function and the cosine function are used to calculate the circRNA Gaussian interaction profile kernel similarity, circRNA functional similarity, miRNA Gaussian interaction profile kernel similarity, and miRNA functional similarity. In order to reduce the influence of noise and redundant information in known interactions, this model uses network embedding to extract the potential feature vectors of circRNA and miRNA, respectively. Finally, an improved inductive matrix completion algorithm based on the feature vectors of circRNA and miRNA is used to identify potential interactions between circRNAs and miRNAs. The 10-fold cross-validation experiment is utilized to prove the predictive ability of the IIMCCMA. The experimental results show that the AUC value and AUPR value of the IIMCCMA model are higher than other state-of-the-art algorithms. In addition, case studies show that the IIMCCMA model can correctly identify the potential interactions between circRNAs and miRNAs.

A Fuzzy System Classification Approach for QSAR Modeling of α- Amylase and α-Glucosidase Inhibitors.

This report proposes the application of a new Machine Learning algorithm called Fuzzy Unordered Rules Induction Algorithm (FURIA)-C in the classification of druglike compounds with antidiabetic inhibitory ability toward the main two pharmacological targets: α-amylase and α-glucosidase. The two obtained QSAR models were tested for classification capability, achieving satisfactory accuracy scores of 94.5% and 96.5%, respectively. Another important outcome was to achieve various α-amylase and α-glucosidase fuzzy rules with high Certainty Factor values. Fuzzyrules derived from the training series and active classification rules were interpreted. An important external validation step, comparing our method with those previously reported, was also included. The Holm's test comparison showed significant differences (p-value<0.05) between FURIA-C, Linear Discriminating Analysis (LDA), and Bayesian Networks, the former beating the two latter according to the relative ranking score of the Holm's test. From these results, the FURIA-C algorithm could be used as a cutting-edge technique to predict (classify or screen) the α-amylase and α-glucosidase inhibitory activity of new compounds and hence speed up the discovery of new potent multi-target antidiabetic agents.

Investigating functional relationships among health and fertility traits in dairy cows

The association between health and pregnancy per artificial insemination, maintenance of pregnancy, and interval to pregnancy in dairy cattle is well documented. However, the understanding of the causal relationships among health and fertility traits is limited. Although information on heritabilities and genetic correlations between traits is key for the development of multiple-trait selection strategies, only knowledge about functional and causal relationships among them allow the prediction of the consequences of external interventions such as management decisions. Thus, this research was carried out to investigate potential causal relationships among multiple health events (retained fetal membranes [RFM], metritis [MET], clinical endometritis [CE], lameness [LS]) and fertility indicators including resumption of ovarian cyclicity (CY) by 50 days in milk and pregnancy on day 60 (P60) after the first artificial insemination in Holstein cows. Mixed-effects structural equation models (SEM) were fitted conditionally on a causal structure among traits, represented by a directed acyclic graph, inferred from the data using the inductive causation (IC) algorithm. The posterior mean and standard deviation of heritability for RFM, MET, CE, LS, CY and P60 were 0.20 ± 0.07, 0.09 ± 0.02, 0.10 ± 0.04, 0.35 ± 0.04, 0.15 ± 0.03, and 0.10 ± 0.03, respectively. The moderate estimates of heritability found for CY and P60 suggest that selective breeding could be effective for the genetic improvement of these traits. The estimates of genetic correlation among traits ranged from -0.82 ± 0.07 (MET and CY) to 0.70 ± 0.13 (RFM and MET). Biological considerations and results from the inductive causation algorithm analysis retrieved six potential causal connections between traits: RFM→MET; RFM→CE; LS→CE; CE→P60; CY→P60; and CY→CE. In addition to the data-driven network recovered, some additional connections or variations on their directions were also considered on the SEM analyses based on biological and chronological knowledge. Overall, the SEM results were very similar to those obtained from the traditional multi-trait mixed model analysis. The advantage of the SEM analysis is that it sheds some light onto the potential causal relationships between traits, which can aid better decisions to manage events occurring during the postpartum of dairy cattle.

Landslide susceptibility mapping at sin Ho, Lai Chau province, Vietnam using ensemble models based on fuzzy unordered rules induction algorithm

Landslide susceptibility map is considered as one of the important steps in assessing vulnerability of an area to landslide hazard. In this study, the main objective is to propose ensemble machine learning models: BF, DF and RSSF which are a combination of Fuzzy Unordered Rules Induction algorithm (F) and three optimization techniques namely Bagging, Decorate, and Random Subspace, respectively for landslide susceptibility mapping. In addition, two other single models namely F and Support Vector Machines (SVM) were also applied for the comparison of performance of the proposed models. For this purpose, the Sin Ho district, Lao Cai Province, Vietnam was selected as the study area. For the development of models, database of 850 present and historical landslides of this province including ten landslide affecting input parameters namely slope, curvature, elevation, aspect, Topographic Wetness Index (TWI), deep division, river density, fault density, aquifer, and geology were used. Validation of the models was done using various popular statistical indicators including Area Under the Receiver Operating Characteristics (AUC) curve. The results show that the BF model (AUC = 0.923) is the best model for accurate landslide susceptibility mapping (LSM) in comparison to other models namely DF (AUC = 0.899), RSSF (AUC = 0.893), SVM (AUC = 0.840), and F (AUC = 0.862). The study revealed that LSM map constructed using BF model can be used for better land use planning and proper landslide hazard management.

Atrial tattoos lines: a new method to terminate atrial fibrillation

Abstract Introduction Atrial fibrillation (AF) is the most common cardiac arrhythmia affecting more than 4% of the world's population. Furthermore, the vulnerable substrate underlying atrial cardiomyopathy remains far from being understood. Atrial cardiomyopathy is a highly complex and heterogeneous disease that occurs in the context of various clinical backgrounds, which explains the heterogenous success of established atrial ablation strategies such as pulmonary vein isolation. Here we developed a new interventional method for the treatment of AF using a tattooing procedure of silver nanowires in the atria to create circular lines of increased conductivity to prevent irregular electrical propagation. Purpose In a translational study, we tested a newly developed interventional method to terminate AF by injecting lines of silver nanowires to increase tissue conductivity. Methods The developed tattoo method was tested in vivo using an established large animal AF model of the domestic pig. Pigs (n=10) with atrial fibrillation were either treated with right atrial tattoo lines using a transmural injection procedure of silver nanowires as closed lines in the atrium or got a sham treatment with saline injections. Over 21 days, AF was induced and monitored (AF burden) by an intracardiac dual-chamber device with a biofeedback induction algorithm. Initially and before final termination, conventional electrophysiological investigation and atrial 3D mapping, echocardiography and epicardial multi electrode array (MEA) measurements were performed. Following the 21 days observation period, the heart was extracted and freshly isolated atrial cardiomyocytes were subjected to cellular electrophysiological, molecular and histochemical characterization. Electrophysiological data was used for in silico modelling of arrhythmia termination and wave propagation. The study protocol was approved by the animal ethics committee. Results The AF burden was significantly reduced (AF burden &amp;lt;10%) in the group treated with the new atrial tattoo line method compared to sham treated pigs. The bi-atrial diameters, quantified by echocardiography, were significantly smaller in the treatment group. Atrial refractory period was significantly shorter in the sham treated pigs. A significant increase of connexin 43 was observed in the injection area of silver nanowires in the myocardium. Measurements with MEA demonstrated increased conduction velocity by a factor of 1.5–2.0 in the areas of silver nanowire injections. Additionally, in silico modelling showed the termination of AF via the created conduction lines. Conclusion The newly developed interventional method of the creation of atrial lines with increased tissue conductivity using silver nanowires could successfully terminate AF in pigs. Perspectively, this new myocardial tattooing technique may be a promising treatment strategy for patients with complex atrial cardiomyopathies to terminate atrial arrhythmias. Funding Acknowledgement Type of funding sources: Foundation. Main funding source(s): Else Kröner Fresenius Foundation

APD: Learning Diverse Behaviors for Reinforcement Learning Through Unsupervised Active Pre-Training

Unsupervised pre-training in reinforcement learning enables the agent to gain prior environmental knowledge, which is then fine-tuned in the supervised stage to quickly adapt to various downstream tasks. In the absence of task-related rewards, pre-training aims to acquire policies (i.e., behaviors) that generate different trajectories to explore and master the environment. Previous research categorizes states into their associated behaviors by learning a supervised discriminator. However, an underlying problem persists: such discriminator is trained in lack of relevant data, leading to an underestimation of reward for new states and inadequate exploration. To this end, we introduce an unsupervised active pre-training algorithm for diverse behavior induction (APD). We explicitly characterize the behavior variables with a state-dependent sampling method, and the agent can decompose the entire state space into parts for fine-grained and diverse behavior learning. Specifically, a particle-based entropy estimator is applied to optimize a combination of behavioral entropy and mutual information objective. Moreover, we develop behavior-based representation learning to compress states into the latent space. Experiments show that our method can improve exploration efficiency and outperforms most state-of-the-art unsupervised algorithms on a number of continuous control tasks in the DeepMind Control Suite.

Security Investment in Cyber-Physical Systems: Stochastic Games With Asymmetric Information and Resource-Constrained Players

This article considers remote state estimation in cyber-physical systems (CPSs) with multiple sensors. Each plant is modeled by a discrete-time stochastic linear system with measurements of each sensor transmitted to the corresponding remote estimator over a shared communication network when their securities are interdependent due to network-induced risks. A dynamic nonzero-sum game with asymmetric information is formulated in which each sensor subject to a resource budget constraint needs to decide whether to invest in security for sending data packets, taking the behaviors of other sensors into account. To overcome the difficulty in characterizing or computing the Nash equilibria (NE), the game with asymmetric information is transformed into another game with symmetric information such that the equilibrium of the original game can be obtained by solving the equilibrium of the new game. Under certain conditions, we devise a backward induction algorithm to obtain a subclass of NE of the original game, known as common information-based Markov perfect equilibria (CIBMPE). Finally, a numerical example is provided to illustrate the results obtained.

Adaptive Ontology-Based IoT Resource Provisioning in Computing Systems

The eagle expresses of cloud computing plays a pivotal role in the development of technology. The aim is to solve in such a way that it will provide an optimized solution. The key role of allocating these efficient resources and making the algorithms for its time and cost optimization. The approach of the research is based on the rough set theory RST. RST is a great method for making a large difference in qualitative analysis situations. It's a technique to find knowledge discovery and handle the problems such as inductive reasoning, automatic classification, pattern recognition, learning algorithms, and data reduction. The rough set theory is the new method in cloud service selection so that the best services provide for cloud users and efficient service improvement for cloud providers. The simulation of the work is finished at intervals with the merchandise utilized for the formation of the philosophy framework. The simulation shows the IoT services provided by the IoT service supplier to the user are the best utilization with the parameters and ontology technique.

Incentive Stackelberg game for H∞$H_\infty$‐constrained multi‐hierarchy systems under observation information

This paper considers the incentive feedback Stackelberg game with multi-hierarchy players under a H ∞ $H_\infty$ constraint, with the proposed solution involving nested hierarchies A and B. In hierarchy A, P0 represents the leader (the leader's control input corresponding to i-th follower, i = 1 , 2 , … , n $i=1,2,\ldots ,n$ ), and P 1 , … , P n $P_1,\ldots ,P_n$ are the followers, with non-cooperative followers induced to virtually cooperate in achieving team-optimal solution and the Nash equilibrium. In hierarchy B, the external disturbance represents the follower, and hierarchy A represents the leader. The main contributions of this work are three-fold. First, an incentive Stackelberg strategy set under the H ∞ $H_\infty$ constraint through observation information is obtained for the first time. Second, a novel iterative algorithm to solve the coupled backward and forward Riccati equations is introduced and thus obtaining an explicit expression of a team-optimal feedback Stackelberg strategy set with an H ∞ $H_\infty$ constraint. Finally, the necessary and sufficient conditions for the existence and uniqueness of the hierarchical game's optimal solutions using the Lyapunov equation and the induction algorithm are presented.

Discovering epistasis interactions in Alzheimer's disease using deep learning model

Alzheimer's disease (AD) is the most common form of dementia. Single Nucleotide Polymorphisms (SNPs) are single nucleotide alterations that can be used as genomic markers disclosing susceptibility to complex diseases like AD. Epistasis has long been significant for recognizing the function of genetic pathways and the evolutionary dynamics of difficult genetic systems. Discovering epistasis interactions holds a vital key to personalized medicine (PM). PM needs a better understanding of the relationship between human genetic data and complex diseases. In this proposed work, a deep neural network (DNN) is applied using SHapley Additive exPlanations (SHAP) to get top 20, 100, 300, and 500 ranking SNPs responsible for AD risk through epistasis interactions. Multi-locus interaction analysis is performed on these identified SNPs using Multifactor Dimensionality Reduction (MDR). This constructive induction algorithm is integrated with DNN for discovering epistasis interactions in a computationally effective method. The proposed framework is applied to Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. The best accuracies are achieved using the top 500 SNPs, and the classification accuracies varied between 0.860 and 0.874 in the five-way interaction model. However, the classification accuracies of 2-way, 3-way, 4-way models varied between 0.663 and 0.670, 0.718 and 0.727, and 0.793 and 0.803, respectively. The results revealed that the reported accuracy scores of the proposed framework outperform the referenced literature work. The proposed framework presents high-ranked risk genes and promising epistasis interactions that may help in explaining the risk of AD.

Novel Evidence-Based Labor Induction Algorithm Associated with Increased Vaginal Delivery within 24 Hours.

This study aimed to assess whether concordance with our proposed labor induction algorithm is associated with an increased rate of vaginal delivery within 24 hours. We conducted a retrospective review of 287 induction of labors (IOLs) at a single urban, tertiary, academic medical center which took place before we created an evidence-based IOL algorithm. We then compared the IOL course to the algorithm to assess for concordance and outcomes. Patients age 18 years or over with a singleton, cephalic pregnancy of 366/7 to 420/7 weeks' gestation were included. Patients were excluded with a Bishop's score >6, contraindication to misoprostol or cervical Foley catheter, major fetal anomalies, or intrauterine fetal death. Patients with 100% concordance were compared with <100% concordant patients, and patients with ≥80% concordance were compared with <80% concordant patients. Adjusted hazard ratios (AHRs) were calculated for rate of vaginal delivery within 24 hours, our primary outcome. Competing risk's analysis was conducted for concordant versus nonconcordant groups, using vaginal delivery as the outcome of interest, with cesarean delivery (CD) as a competing event. Patients with 100% concordance were more likely to have a vaginal delivery within 24 hours, n = 66 of 77 or 85.7% versus n = 120 of 210 or 57.1% (p < 0.0001), with an AHR of 2.72 (1.98, 3.75, p < 0.0001) after adjusting for delivery indication and scheduled status. Patients with 100% concordance also had shorter time from first intervention to delivery (11.9 vs. 19.4 hours). Patients with ≥80% concordance had a lower rate of CD (11/96, 11.5%) compared with those with <80% concordance (43/191 = 22.5%; p = 0.0238). There were no differences in neonatal outcomes assessed. Our IOL algorithm may offer an opportunity to standardize care, improve the rate of vaginal delivery within 24 hours, shorten time to delivery, and reduce the CD rate for patients undergoing IOL. · Studies on IOL have focused on individual steps. A labor induction algorithm allows for standardization.. · Algorithm concordance is associated with decreased time to delivery.. · Algorithm concordance is associated with decreased CD rate..

Modeling landslide susceptibility using data mining techniques of kernel logistic regression, fuzzy unordered rule induction algorithm, SysFor and random forest

Modeling landslide susceptibility using data mining techniques of kernel logistic regression, fuzzy unordered rule induction algorithm, SysFor and random forest

An inductive heterogeneous graph attention-based multi-agent deep graph infomax algorithm for adaptive traffic signal control

Adaptive traffic signal control (ATSC) facilitates alleviating traffic congestion. Multi-agent deep reinforcement learning (MDRL) is a new promising algorithm for ATSC, and Graph Neural Networks (GNNs) further promote its learning ability. However, there are some drawbacks in the state-of-the-art MDRL algorithms. (1) These algorithms cannot effectively fuse diverse heterogeneous information of traffic networks due to adopting homogeneous GNNs; (2) These algorithms cannot be effectively trained due to merely adopting MDRL loss functions. In this paper, we propose an Inductive Heterogeneous graph Attention-based Multi-agent Deep Graph Infomax (IHA-MDGI) algorithm for ATSC. The proposed IHA-MDGI algorithm conducts both feature fusion via a proposed Inductive Heterogeneous graph Attention (IHA) algorithm and training via a proposed Multi-agent Deep Graph Infomax (MDGI) framework. Specifically, (1) Unlike the algorithms which adopt homogeneous GNNs, in the IHA algorithm, heterogeneous GNNs are designed to fuse both heterogeneous structural information and heterogeneous features of traffic networks, which aims to acquire heterogeneous information embeddings of traffic networks. (2) In the MDGI framework, the acquired embeddings are used to calculate the signal-control policies and Q-value for each agent, and then a mutual-information loss function is designed, which combines with the MDRL loss function to jointly train the whole algorithm. The designed mutual-information loss function focuses on maximizing mutual information between input (i.e., heterogeneous information embeddings) and output (i.e., Q-value), which can produce cooperative signal-control policies and maximize Q-value. We conduct the experiments in both real-traffic and synthetic-traffic networks under the time-varying traffic flows, and the results demonstrate that IHA-MDGI algorithm outperforms the state-of-the-art MDRL algorithms about multiple metrics.