The heuristic are approximation methods for hard problems and for contexts with hard constraints of reactivity (so when exact methods are too time consuming). The metaheuristics are heuristics patterns : you have to adapt it for your problem.
About the multi-heuristics, I'm not sure: I guess the process is to use simultaneously different heuristics well adapted for the problem, and finally for keeping the best solution. Maybe someone else could explain it?
The heuristic is usually some simple methods motivated by some intuitive characters, such as local search that only accepts improving solution, or some constructing method: insertion or saving method for VRP.
Meta-heuristic is more sophisticated method, which use some strategies to lead the search to escape from the local optimal. These methods includes Simulated Annealing, Tabu Search, VNS, ACO, GA.
Multi-heuristic usually contains two levels, the lower level contains several simple heuristics, and the high level is used to guide the lower level heuristics.
Heuristic is a local problem solving technique, while meta heuristic takes a global view of the problem. Multi heuristic combines or selects from different heuristics to solve the problem at hand. It is a more universal version where different techniques can be combined to form an hybrid solution technique. Such a technique may be used to solve problems from different domains.
Heuristics are problem specific, meta heuristics are generalized heuristics. Examples of meta heuristics are GA,SA,ACO,TS etc.
Multiheuristics are useful when you decompose the problem into several levels and each level we can use some heuristics to solve. Basically you need combination of heuristics to solve the problem that is why they call it as multiheuristics.
Optimization algorithms can be roughly divided into two categories: exact algorithms and heuristics. Exact algorithms are designed in such a way that it is guaranteed that they will find the optimal solution in a finite amount of time. However, for very difficult optimization problems (e.g. NP-hard or global optimization) this "finite amount of time" may increase exponentially respect to the dimensions of the problem. Heuristics do not have this guarantee, and therefore generally return solutions that are worse than optimal. However, heuristic algorithms usually find "good" solutions in a "reasonable" amount of time.
Many heuristic algorithms are very specific and problem-dependent. On the other hand, a metaheuristic is a high-level problem-independent algorithmic frame-work that provides a set of guidelines or strategies to develop heuristic optimization algorithms. But a concrete definition has been elusive and in practice many researchers and practitioners interchange these terms. Thus, the term metaheuristic is also used to refer to a problem specific implementation of a heuristic optimization algorithm according to the guidelines expressed in such a framework.
You have also mentioned the optimum. These stochastic techniques do not guarantee finding an optimum. Near optimum or very poor solutions can be found. However, a solution can be returned at any time.
So it is important to have many runs to establish the quality of your heuristics of metaheuristics.
In contrast to heuristics, meta-heuristics designate a computational method that optimizes a problem by iteratively trying to improve a candidate solution with regard to a given measure of quality. It can search very large option spaces of candidate solutions.
A metaheuristics is an algorithm designed to solve approximat ely a wide range of hard optimization problems without having to deeply adapt to each problem. Indeed, the greek prefix ‘‘meta’’, present in the name, is used to indicate that these algorithms are ‘‘higher level’’ heuristics , in contrast with problem-specific heuristics. Metaheuristics are generally applied to problems for which there is no satisfactory problem-specific algorithm to solve them ( Boussaïd et al.(2013).
Boussaïd, I., Lepagnot, J., & Siarry, P. (2013). A survey on optimization metaheuristics. Information Sciences, 237, 82-117.