Short-term road traffic speed prediction is a long-standing topic in the area of Intelligent Transportation System. Apparently, effective prediction of the traffic speed on the road can not only provide timely details for the navigation system concerned and help the drivers to make better path selection, but also greatly improve the road supervision efficiency of the traffic department. At present, some researches on speed prediction based on GPS data, by adding weather and other auxiliary information, using graph convolutional neural network to capture the temporal and spatial characteristics, have achieved excellent results. In this paper, the problem of short-term traffic speed prediction based on GPS positioning data is further studied. For the processing of time series, we innovatively introduce Dynamic Time Warping algorithm into the problem and propose a Long Short-Term Memory with Dynamic Time Warping (D-LSTM) model. D-LSTM model, which integrates Dynamic Time Warping algorithm, can fine-tune the time feature, thus adjusting the current data distribution to be close to the historical data. More importantly, the fine-tuned data can still get a distinct improvement without special treatment of holidays. Meanwhile, considering that the data under different feature distributions have different effects on the prediction results, attention mechanism is also introduced in the model. Our experiments show that our proposed model D-LSTM performs better than other basic models in many kinds of traffic speed prediction problems with different time intervals, and especially significant in the traffic speed prediction on weekends.

Physical--observation dual-axis (PODA) Theory holds that physical reality is realized jointly through physical state and observation state. A change of observation state can therefore change the realized physical fact even when the physical state is held fixed. We develop the phase law of this PODA structure. A coherent readout of the joint observation law gives a smooth scalar complex response \( g \). Wherever this response is nonzero, differentiating its polar decomposition gives the observed phase rate and the one-form \( A_g=\operatorname{Im}(\mathrm d g/g) \). A transported phase with the same increments differs from the observed phase by a constant reference factor. We derive its differential equation and unique initial-value solution, and show that the equation is horizontal lifting for the unique response-compatible connection on the trivial principal \( \mathrm U(1) \) phase bundle. We call the resulting equation \( \widetilde c^{\,*}\Omega_g=0 \), the Phase Transport Fundamental Equation (PTFE), the first fundamental equation of observation space in PODA Theory. Its group-valued, covariant-derivative and continuous real-phase forms express one transport law: a path and an initial phase determine a unique horizontal lift. Integration gives finite phase comparison, composition along successive paths, and inverse transport between two observation states of the same physical source. Positive amplitude rescaling leaves the law unchanged; synchronized changes of response and fiber coordinates preserve its covariant form. The induced connection is flat and has trivial circular holonomy, although a continuously unwrapped phase may wind. We distinguish the full response \( \mathsf G_0, \) the phase-transport structure \( \mathsf G_1 \), and the task quotient \( \mathsf G_2 \). At the last level, maximal invariants describe the quotient, and a target can be recovered exactly when it is constant on each fiber of the quotient observation. Phase transport and the information retained after a task reduction thus follow from the same joint observation law.

An observation record may depend both on the physical state assigned to the object and on the state through which the record is formed. We give a self-contained set-theoretic derivation of Physobser Axiom and ask when the latter dependence can be removed without changing the declared physical-state semantics. We begin with an empirically typed, task-relative closedworld table µ0 : P × A → Y⊥, where ⊥ ∈ Y / denotes certified nonimplementability rather than missing data. Complete row and column behavior induce quotient maps qX : P ↠ X and qS : A ↠ S, a unique totalized law Fe : X × S → Y⊥, and the admissible domain B = Fe−1 (Y). We prove that the two-sided quotient is terminal among surjective typed exact factorizations and that every biextensional factorization is uniquely isomorphic to it. When an antecedent physical theory supplies a compatible surjection r : P ↠ Xphys, physical identity is retained and only the observation-forming side is minimized. The resulting S is the terminal nonredundant observer-state realization and is unique up to a unique behavior-preserving isomorphism. The Set-Theoretic Physobser Representation Theorem identifies fixed-physicalstate formation non-triviality, |S| ≥ 2, and failure of the totalized law to factor through the physical projection as equivalent statements; two unequal implementable outputs yield the corresponding obstruction for F : B → Y. These results are compressed into Physobser Axiom, an ordered physical–observation interface with a physical-state axis and an observerstate axis. Within Physobser Theory, this axiom is the foundational structural statement, and the resulting dual-axis interface is a basic mathematical structure of the theory rather than its complete content. Kernel relations then give exact criteria for conditional, label-free, and joint identifiability, while B ×Xphys B is the canonical same-source comparison domain. The derivation uses sets, maps, equivalence relations, quotients, factorizations, and fiber products alone.

Classical theories typically model a complete record as a unary function of a physical state, treating observation conditions as external parameters. However, quantum mechanics exhibits a fundamental tension: the same independently certified physical input can yield different complete records when observation conditions vary, giving rise to the century-old measurement problem and the EPR paradox. In this work, we derive the necessary state architecture to resolve this without conflating the physical input with the conditions through which its record is formed. By applying behavioral equivalence and redundancy elimination within the category of sets, we prove that the formation-side difference cannot be mathematically removed. This forces the “observation state” to emerge as an irreducible, objective, state-bearing axis, establishing a strict physical–observation dual-axis (PODA) architecture. We emphasize that this dual-axis structure represents two non-interchangeable projection responsibilities, rather than orthogonal coordinates or dynamically decoupled systems. Under stated coherence and composition conditions, this dual-axis geometry naturally yields the complex field, binary projective geometry, and the Born trace pairing. Crucially, it further derives the Schrödinger equation as a covariant temporal composition, the CHSH inequality, and the attainable Tsirelson bound. In standard quantum mechanics, the complex field, the Schrödinger equation, and the Born rule are independent axioms. In our framework, they are unified as geometric corollaries of the PODA structure. Within this framework, the notorious quantum paradoxes are naturally dissolved: “wave-function collapse” is revealed as a conditional transition at the observation interface, and “decoherence” is understood as the objective geometric coupling between the physical and observation axes, rather than random environmental noise. Bell violation is thereby identified not as superluminal influence, but as the absence of a positive global answer section across incompatible observation states. Finally, the theory proposes a specific table-top experiment that predicts a sharp, parameter-free numerical constant: a geometric phase of 0.7π ≈ 2.1991 radians. This value directly distinguishes the dualaxis mechanism from standard quantum mechanics, which predicts exactly zero for the same configuration. Observing this constant would confirm the physical reality of the observation axis, whereas a null result falsifies the mechanism.

Whether observation dynamics can leave distinct matter records under identical physical initial conditions and subsequent external forcing depends on three questions: whether observation initial states have a well-defined identity, whether they can be prepared, and whether their evolution satisfies the common field equations. We construct a local realization within the existing five-dimensional response space and quadratic common action of PODA. With spatial and temporal references fixed, primitive response transport yields nonzero mixed curvature and identifies the observation coordinate through distinguishable comparison records; the image of its field data is an invariant subspace of the transverse dynamics. The field restoring force and a nonzero current overlap allow a single matter input to prepare arbitrary sufficiently small joint initial states. An explicit eighth-degree polynomial protocol carries the state initially at rest to two final states with identical matter position, matter velocity, and observation configuration, but opposite observation velocities. The input vanishes continuously at both endpoints, and the net work equals the stored energy. After release, the two-frequency displacement and the observation comparison probability obey a common integral identity. We also derive propagation relations for initial-state mismatch, finite readout precision, and control-waveform errors, and retain the complete calculations for matter sources, the one-loop relativistic response, and indirect tests involving common parameters. The result is an existence theorem for preparation and response in the specified local realization, within the stated quadratic dynamics and external matter-source model.